In H.264/AVC standard, motion estimation can be processed on multiple reference frames (MRF) to improve the video coding performance. The computation is also increased in proportion to the reference frame number. Many software oriented fast multiple reference frames motion estimation (MRF-ME) algorithms have been proposed. However, for the VLSI real-time encoder, the heavy computation of fractional motion estimation (FME) makes the integer motion estimation (IME) and FME must be scheduled in two macro block (MB) pipeline stages, which makes many fast MRF-ME algorithms inefficient. In this paper, one edge gradient detection based algorithm is provided to reduce the computation of MRF-ME. The image being rich of texture and sharp edges contains much high frequency signal and this nature makes MRF-ME essential. Through analyzing the edges' gradient, we just perform MRF-ME on those blocks with sharp edges, so the redundant ME computation can be efficiently reduced. Experimental results show that average 26.43% computation can be saved by our approach with the similar coding quality as the reference software. This proposed algorithm is friendly to hardwired encoder implementation. Moreover, the provided fast algorithms can be combined with other fast ME algorithms to further improve the performance.